Pre-primary polls predicted Democratic Senate candidate Abdul El-Sayed would win over his opponent Haley Stevens by large margins. But instead of a blowout, it was a narrow victory in an extremely close race.
Those results have many people wondering what pollsters got wrong and if polls can do more harm than good to campaigns.
Michigan State University political scientist Corwin Smidt says it's likely pollsters were modelling their electorate using bad or incomplete data compiled by national firms. He has his own personal examples to prove his point.
"My wife, I know voted Democrat. She is listed in the file as voting Republican. So, my mother has voted in the same address in Michigan for 40 years. She is missing from this data set. I don't know why she lives in the same house as my dad. My dad's in the data set. He's listed as living alone," he explained.
Primaries also become difficult to model because voters can be motivated to cast their ballot for a specific candidate for much more varied reasons rather than just party loyalty That combined with those not so accurate data sets may have led pollsters astray, Smidt says.
"Either they were seeing something in how survey respondents were responding to say that the electorate was going to be different, or they were modeling a different election," he said. "What we saw is that they, in many cases, they were modeling a different election—not one that we typically see in Michigan, and not one that we did see this August."
Smidt predicts there won't be as big of a gap between polling and results come November for the general election, just because people vote more reliably and consistently for one of the two parties.
Smidt adds bad polling can have bigger consequences. Candidates often decide to enter races or build their campaigns around issues they think will resonate with voters after consulting polls.
"A lot of these local candidates, if they're using a national, some New York or Washington D.C. pollster who's just cheap, you know, newer, younger, more progressive, more conservative, whatever. They come in and they just say, 'This is what I think Michigan looks like,'" he said.
"It's different than maybe what someone who's been a seasoned pollster in Michigan might say is typically this is what Michigan voters think. So, these candidates are then are running on issues that are not necessarily what people are interested."
Smidt says part of the problem can be traced to reduction in public independent polls done by universities or media companies while private polling firms driven by smaller and smaller political contingents have grown.
"Now we don't even have Democratic firms and Republican firms. We have, you know, Trump Republican firms and not Trump Republican firms, and we have progressive Democrat firms and sort of moderate Democrat firms."
Ultimately, he advises voters to largely ignore polls and to cast their ballot for what will make them happy.
"Trying to be the smart voter is not the way to do things. Likewise, if you think that you might you know sort of be upset by sort of being misled by a poll, you know, it doesn't really benefit to know who's going to win in a lot in a lot of ways," he said.
"It benefits you more to know whether you're going to like the candidate."
Interview Highlights
On where pollsters get their information from
Michigan election data is not that informative. We have so we live in a state with 1,500 separate clerks, a really diffuse election administration system, which is great for some things, but it also makes it hard for the data to sort of have centrality and be understandable to national vendors. And so, a lot of the national data sets that have Michigan in them have Michigan coded either incorrectly or Michigan doesn't quite fit in the national boxes that for sort of August primary election turnout. So, some pollsters recently have posted online, and they said, "Well, turnout in you know August 2022 was this among Democrats." We have no public information that that is true. That is made up.
On if polling for the November general election will be closer to the results
There's this thing called partisanship, which makes things really accurate to predict, and unfortunately, you know, that's American politics being the way it is. Partisanship anchors a lot of responses. So, you know, it's a lot easier to identify, okay, how many voters from Republicans and Democrats are we're going to see? How many voters voted for Trump in 2024 and for Harris in 2024? And you can sort of weigh that out. ... The way we model national in November elections is more consistent across states, and it's a lot easier to do. So, the accuracy will probably be improved.
On if voters should pay attention to polls
I always like saying that your vote never matters, so why not just vote the way you want to think? If people are sitting there trying to be strategic with their vote, they're only going to be unhappy. The vote that makes you happy is the vote that you should make. Trying to be the smart voter is not the way to do things. Likewise, if you think that you might you know sort of be upset by sort of being misled by a poll, you know, it doesn't really benefit to know who's going to win in a lot in a lot of ways. It benefits you more to know whether you're going to like the candidate. There are so many other things to think about that could could help you as a voter or just maybe maybe don't even pay attention to that stuff. Why do you need to know who's going to win? So, those are some things where it's just, you know, as a voter, you can just sort of say, well, why do I need to know this?
Interview Transcript
Sophia Saliby: Pre-primary polls predicted Democratic Senate candidate Abdul El-Sayed would win over his opponent Haley Stevens by large margins. But instead of a blowout, it was a narrow victory in an extremely close race.
Those results have many people wondering what pollsters got wrong and if polls can do more harm than good to campaigns and to voter sentiment.
Corwin Smidt is a Michigan State University political scientist, and he joins us now. Thanks for being here.
Corwin Smidt: Thank you.
Saliby: What do you think pollsters missed when it came to the Democratic Senate race? Was it just one big thing or a bunch of little things?
Smidt: When we talk about pollsters, we're talking about a large group of people, not necessarily who are the same pollsters four years ago or two years ago. These were new firms. This was a hot topic election that was generating a lot of national interest. A lot of pollsters saw an opportunity to come in here and sort of prove their merit or accuracy, and we're trying to get publicity out of this for themselves to say they got the race right.
What they got wrong, I can tell you one clear thing is that Michigan election data is not that informative. We live in a state with 1500 separate clerks, a really diffuse election administration system, which is great for some things, but it also makes it hard for the data to sort of have centrality and be understandable to national vendors.
And so, a lot of the national data sets that have Michigan in them have Michigan coded either incorrectly or Michigan doesn't quite fit in the national boxes for sort of August primary election turnout. So, some pollsters recently have posted online, and they said, "Well, turnout in, you know August 2022 was this among Democrats." We have no public information that that is true. That is made up.
That data set was made up by a vendor nationally who guessed. My wife, I know voted Democrat. She is listed in the file as voting Republican. So, my mother has voted in the same address in Michigan for 40 years. She is missing from this data set. I don't know why she lives in the same house as my dad. My dad's in the data set. He's listed as living alone. And these are the national data files that these pollsters are using. They don't really understand. They just pull it from this data set, and they just generate an estimate of what they think the turnout is going to be. They don't understand that the data in this data set is actually not true. It's made up, and it's just a guess.
So, when they do this, they model an electorate, and you could see it ahead of time, the electorate that they were modeling for Michigan and for the Democratic election was much different than what we had seen in the past, and it was very odd. So, either they were seeing something in how survey respondents were responding to say that the electorate was going to be different, or they were modeling a different election. And what we saw is that they, in many cases, they were modeling a different election—not one that we typically see in Michigan, and not one that we did see this August.
Saliby: You posted on on social media that bad polling can have larger consequences down the line. So, why is bad polling a problem for our elections?
Smidt: Well, if you think about it, candidates before they go run for office, they take polls, they try to see what voters are thinking. If they're using these pollsters who don't have the right data, they have a wrong impression of what Michigan voters are thinking. So, a lot of these local candidates, if they're using a national, some New York or Washington D.C. pollster who's just cheap, you know, newer, younger, more progressive, more conservative, whatever. They come in and they just say, "This is what I think Michigan looks like."
It's different than maybe what someone who's been a seasoned pollster in Michigan might say is typically this is what Michigan voters think. So, these candidates are then are running on issues that are not necessarily what people are interested in. I mean, so we could talk about whether they're too far left or too far right, but I think the bigger issue is actually maybe they're ignoring issues that voters care about, or maybe they're focusing on issues they think are important issues in Michigan that actually Michigan voters don't care that much about.
So, rather than thinking about too far too left, too right, I think the question is: are they identifying and seeing the types of issues that are really on Michigan voter minds, or are they sort of recreating this from poorly weighted survey data that says actually, you know, this issue is the most important when that's just an artifact of the modeling choice that the the pollster made?
Saliby: Is this something that can be fixed, you know, going into November? Are we going to get maybe more accurate polls, or is there no way to know?
Smidt: Thankfully, well, depends how you want to look at it, there's this thing called partisanship, which makes things really accurate to predict, and unfortunately, you know, that's American politics being the way it is, partisanship anchors a lot of responses. So, you know, it's a lot easier to identify, okay, how many voters from Republicans and Democrats are we're going to see? How many voters voted for Trump in 2024 and for Harris in 2024? And you can sort of weigh that out. And it's a lot easier.
There's more common knowledge about what these results look. So, the modeling choices by the pollsters are less important. Now Michigan, being what it is—a battleground, purple state, whatever you want to call it—we've had these very small margins. And while polls haven't necessarily been, especially the local pollsters, the ones who are from Michigan, actually haven't done that poorly in the prior years.
But what we're seeing is there are fewer public polls done by local state media companies, the Detroit Free Press stopped sponsoring a poll this year when they had in the past. So, we have fewer state-run independent polls. So, that means that a lot more of the polls are done by national firms that are being paid for by political groups. And so, we probably will see more of that. However, what you will see is that these groups now are going to be polling in a way that is more sort of accustomed to the national data sets, the way we model national in November elections is more consistent across states, and it's a lot easier to do. So, the accuracy will probably be improved.
Saliby: So, should voters just not pay attention to polls?
Smidt: You know, it's well, okay, so one, I always like saying that your vote never matters, so why not just vote the way you want to think? If people are sitting there trying to be strategic with their vote, they're only going to be unhappy. The vote that makes you happy is the vote that you should make. Trying to be the smart voter is not the way to do things. Likewise, if you think that you might sort of be upset by sort of being misled by a poll, you know, it doesn't really benefit to know who's going to win in a lot in a lot of ways. It benefits you more to know whether you're going to like the candidate.
There are so many other things to think about that could could help you as a voter or just maybe maybe don't even pay attention to that stuff. Why do you need to know who's going to win? So, those are some things where it's just, you know, as a voter, you can just sort of say, well, why do I need to know this?
I think the broader issue here, though, is that actually this is the role that public data and universities and other institutions used to play, and more and more we're becoming more privatized, more balkanized. We're having much more polarization, and so even within Democrat, Republican parties, there's divisions within the party, and now we don't even have Democratic firms and Republican firms. We have, you know, Trump Republican firms and not Trump Republican firms, and we have progressive Democrat firms and sort of moderate Democrat firms.
Now we have even divisions within these groups, and so it's really hard to parse out even, let alone talk about accuracy and whether it's rigorous, it's really hard to parse out. Oh, how was the question asked? What was kind of the topics that were asked in the survey? There's just too much to focus on, that to really understand whether it really is a good indicator of where American, Michigan voters are thinking.
Saliby: Corwin Smidt is a political scientist at MSU. Thank you for joining us.
Smidt: You're welcome. Thank you.
This conversation has been edited for clarity and conciseness.
